Analysis of Teaching Mode Innovation and Learning Effectiveness Assisted by Artificial Intelligence

In recent years, with the rapid development of artificial intelligence technologies such as big data, cloud computing, and deep learning, it has greatly promoted social progress, and its use in daily life and work has become more and more widespread [1]. Under the guidance of national policies, the education field is also exploring the use of AI-enabled education and teaching to promote the transformation of education and teaching from empirical to personalized, intelligent and precise [2-4].

Changes in teaching methods are usually based on relevant teaching elements, and the use of the technological advantages of artificial intelligence can be innovatively integrated in the teaching environment, teaching organization and teaching evaluation [5]. In order to realize the effective integration of online and offline, before and after class, virtual and reality by building an interactive intelligent teaching support system. The intelligent system provides students with accurate learning resources, guides students to explore learning independently, collects learning data in real time, and intelligently measures and evaluates learning effectiveness [6-8]. In the process of teaching implementation, teachers can use the real-time feedback provided by the intelligent system to accurately organize classroom teaching and dynamically optimize teaching strategies.

The use of VR, AR and other technologies can enrich the presentation form of teaching resources, constantly improve the visual and sensory experience of the learner, but also reconfigure the existing learning space, so that the learning space contextualization, virtualization, ubiquitous, breaking the time and space limitations of the original learning space [9]. With the support of artificial intelligence technology, on the one hand, it can provide adapted teaching resources based on the personalized needs of students’ learning and help students obtain a good learning experience. On the other hand, the collection and generation of resources is more efficient, along with the increase of learning participants and the advancement of the learning process, there will be more generative resources, for example, through the intelligent identification technology and intelligent comparison technology, more wrong resources and wrong operations can be generated, and these resources will be more targeted for students’ learning [10-12]. In addition, more process resources can be generated, for example, students’ learning behaviors and processes will be recorded throughout the process, generating data, and through the use of learning analytics, data mining, and other artificial intelligence technologies, the best learning paths can be adapted according to the learning of different learners to achieve the accurate push of teaching resources [13-14]. For example, push according to the key points, push according to the error points, push according to the learning content clicked by students, and push according to the type of resources that students are interested in, etc., which can help learners significantly improve the efficiency of learning [15].

China has continued to promote the construction and improvement of the hardware environment for education informatization, the enhancement of teachers’ and students’ information technology (IT) capabilities, and the integration of IT with the curriculum, with remarkable results [16]. The Ministry of Education and education management departments at all levels have implemented a series of important initiatives in the construction of resource service system, the construction and application of network learning space, and the construction of artificial intelligence-assisted teacher team to comprehensively promote the integration of emerging information technology and education represented by artificial intelligence [17-18]. Being in an intelligent technological environment with rapid technology iteration, intelligent teaching software and hardware environment, data fusion drive, and large platform aggregation application, we start from the technological environment in which the teachers are located, analyze the composition of the technological environment and the new requirements of the environmental changes on the teachers, and make it clear that the focus point of teachers’ teaching and learning innovations under the support of the environment will provide a directional reference for the teachers to take the initiative of innovation and application, and to achieve the in-depth fusion of the information technology and teaching in the new era. [19-21].

In the research on the integration of AI and education, some scholars focus on the role played by AI technology in the field of education, such as data collection and analysis, intelligent assessment, intelligent tutoring, and personalized teaching, etc. Literature [22] systematically analyzes the research and practice of AI-enabled education, focusing on the analysis of intelligent collaborative learning, intelligent tutoring, intelligent teaching assessment, and personalized teaching, and affirms the superiority of AI technology in the field of education. Literature [23] describes that AI technology empowers educational data collection, effectively improving the consistency, validity and uniqueness of data collection. Literature [24] analyzed the development path of AI in education based on learners, educators and teaching systems, and focused on the personalized adaptive teaching model based on AI technology, and concluded that AI technology significantly promotes the innovation of education.

However, more scholars study and analyze the role played by AI technology plus education and the logical mechanism, literature [25] in the form of intelligent voice and image interaction observation, revealing that the theory of multiple intelligences and AI technology can provide teachers with effective intelligent teaching aids, which can help personalized teaching and the cultivation of composite innovative talents. Literature [26] discusses the role and impact played by artificial intelligence in the field of education, and explores the innovative development path of artificial intelligence technology in education, in order to promote the construction of talent cultivation mode adapted to the development of artificial intelligence. Literature [27] carries out a controlled teaching experiment to demonstrate that the flipped classroom and personalized teaching video recommendation mode based on artificial intelligence technology can improve students’ learning performance more than the traditional teaching mode. Literature [28], based on the questionnaire survey method, points out that AI-assisted teaching, assisted practice, assisted examination, assisted evaluation has been developed and mature, and has a positive effect on the improvement of teaching effect, in which the teachers’ cognition of the education of AI technology is positively correlated with the impact of the AI technology-enabled education. Literature [29], based on the data collected from the questionnaire survey and the correlation analysis of least squares structural equation modeling, pointed out that the AI competence in colleges and universities consists of three formative second-order variables, namely, resources, skills, and awareness, and that the AI intelligence in colleges and universities is positively correlated with students’ self-efficacy, creativity, and learning performance.

Finally, some experts in the field of education are also thinking and exploring the path of integrating AI technology into education and promoting educational innovation and reform based on AI education research and experimental methodology, literature [30] after a 16-week environmental education based on AI technology as the infrastructure, students’ environmental knowledge and ability have been improved to a certain extent, and the study has deepened the understanding of the effect of AI technology used in teaching and learning, and contributed to the innovation of smart education. The study deepens the understanding of the effectiveness of AI technology for teaching and helps to innovate and optimize smart education. Literature [31] used controlled experiments to confirm that AI teaching strategies have a positive effect on students’ performance and that students’ attitudes towards AI technology play a mediating role, which is statistically significant.

This paper proposes a deep knowledge tracking model (DKT-CF) incorporating the Ebbinghaus forgetting curve, aiming to take the effects of forgetting behavior into account in the model. The model constructs a forgetting module that calculates a learner’s memory retention of knowledge at a given moment by modeling the Ebbinghaus forgetting curve. Combined with behavioral characteristics that may influence forgetting, forgetting factors were constructed and spliced with interaction sequences as inputs to the model, which were then predicted using item response theory. Comparison experiments with the GKT, AKT, and CoKT benchmark models were conducted on three datasets: ASSIST10, ASSIST15, and EdNet. After that, we designed and developed an intelligent teaching and learning system, and constructed an innovative teaching model based on STSE on this basis, and finally verified the effectiveness of the model through covariance analysis and correlation analysis.

2

DKT-FC based intelligent teaching aid innovative teaching mode

2.1

DKT-FC Modeling Framework

The Deep Knowledge Tracking Model Fusing Forgetting Curves [32-33] (DKT-FC) splices the forgetting factor with the learner’s interaction information at the input layer to obtain a sequence of interactions containing forgetting behaviors, which serves as an input to the LSTM network. The LSTM network can efficiently process such sequence data by capturing the temporal dependencies and nonlinear relationships in the sequence, thus more accurately inferring the learner’s knowledge state.

2.1.1

Oblivion module

According to classical theories in the field of educational psychology, especially Ebbinghaus’ theory of the forgetting curve and the decline of memory traces, learners’ forgetting behavior is affected by a combination of multiple factors. Among them, the repetition interval, the learning interval, and the number of repetitions are considered to be the main factors affecting forgetting behavior. These factors work together in the learner’s memory system to determine the retention and forgetting rate of what they have learned. In this study, a forgetting module was constructed and the Ebbinghaus forgetting equation was applied to quantify the memory retention of knowledge.

The difference between the last time the point was learned and the current learning time was calculated to obtain the repetition interval Δt (in minutes), which was substituted into the Ebbinghaus forgetting equation: 1 $X = \frac{1.84}{{(\log_{10} Δ t)}^{1.25} + 1.84}$ $$X = \frac{{1.84}}{{{{\left( {{{\log }_{10}}\Delta t} \right)}^{1.25}} + 1.84}}$$

To derive the learner’s current memory retention of the knowledge point X. Next, the learning interval, the number of repetitions, and the memory retention rate are uniquely hot coded and these coded features are subjected to a splicing operation to ultimately obtain the forgetting factor G. Viz: 2 $G = [X \oplus Δ T \oplus n]$ $$G = \left[ {X \oplus \Delta T \oplus n} \right]$$

where ⊕ denotes a splice operation.

2.1.2

Input layer

In this paper, we use the splicing operation to combine the learner’s forgetting factor G and the interaction sequence x_t into a new feature, so that the model can take into account the effect of forgetting behavior on the answer results according to the principle of pedagogy, and then improve the prediction accuracy of the model. The splicing operation is as follows: 3 $F = [G \oplus x_{t}]$ $$F = \left[ {G \oplus {x_t}} \right]$$

2.1.3

LSTM

In order to accurately capture the learner’s mastery state of knowledge, the Long Short-Term Memory Network [34] (LSTM) was used to model the whole exercise process in depth. Through the processing of LSTM, the model is not only able to capture the sequential dependency between each exercise, but also able to fully consider the potential correlation and interactions between exercises. 4 $i_{t} = σ (W_{i} [e_{t}, h_{t - 1}, c_{t - 1}] + b_{i})$ $${i_t} = \sigma \left( {{W_i}\left[ {{e_t},{h_{t - 1}},{c_{t - 1}}} \right] + {b_i}} \right)$$ 5 $f_{t} = σ (W_{f} [e_{t}, h_{t - 1}, c_{t - 1}] + b_{f})$ $${f_t} = \sigma \left( {{W_f}\left[ {{e_t},{h_{t - 1}},{c_{t - 1}}} \right] + {b_f}} \right)$$ 6 $o_{t} = σ (W_{o} [e_{t}, h_{t - 1}, c_{t - 1}] + b_{o})$ $${o_t} = \sigma \left( {{W_o}\left[ {{e_t},{h_{t - 1}},{c_{t - 1}}} \right] + {b_o}} \right)$$ 7 $c_{t} = f_{t} c_{t - 1} + i_{t} \tanh (W_{c} [e_{t}, h_{t - 1}] + b_{c})$ $${c_t} = {f_t}{c_{t - 1}} + {i_t}\tanh \left( {{W_c}\left[ {{e_t},{h_{t - 1}}} \right] + {b_c}} \right)$$ 8 $h_{t} = o_{t} \tanh (c_{t})$ $${h_t} = {o_t}\tanh \left( {{c_t}} \right)$$

where i_t, f_t, o_t, c_t, and h_t, denote the input gate, oblivion gate, output gate, cell state, and hidden state, respectively.

2.1.4

Forecasting layer

IRT (Item Response Theory) [35], as a fundamental theory in the field of statistical psychology, has demonstrated excellent interpretability in the application of knowledge tracking. In this study, a two-parameter IRT model containing three key parameters, which are item discrimination α, learner ability θ, and problem difficulty β, is used. The parameters of the two-parameter IRT model in this section are as follows.

Item Distinctiveness: 9 $α = σ (W_{α} \cdot x_{t} + b_{α})$ $$\alpha = \sigma \left( {{W_\alpha } \cdot {x_t} + {b_\alpha }} \right)$$

Learner competencies: 10 $θ = \tanh (W_{θ} \cdot h_{t} + b_{θ})$ $$\theta = \tanh \left( {{W_\theta } \cdot {h_t} + {b_\theta }} \right)$$

Problem Difficulty: 11 $β = \tanh (W_{β} \cdot d_{t} + b_{β})$ $$\beta = \tanh \left( {{W_\beta } \cdot {d_t} + {b_\beta }} \right)$$

The IRT item response theory was used to predict the learner’s future performance in responding using the formula: 12 $P (y_{i j} = 1 | θ_{i}) = \frac{1}{1 + e^{- D a_{j} (θ_{i} - β_{i})}}$ $$P\left( {{y_{ij}} = 1|{\theta_i}} \right) = \frac{1}{{1 + {e^{ - D{a_j}\left( {{\theta_i} - {\beta_i}} \right)}}}}$$

2.2

Validity of the DKT-FC model

2.2.1

Performance comparison

In order to verify the effectiveness of the proposed DKT-FC model, this paper conducts experiments on three datasets and compares the performance with three representative methods. In this paper, the embedding size of the practice answer interactions and the hidden state size of the LSTM are both set to 90. The maximum learning length sequence is set to 60, and if the learning sequence is longer than 60, it is divided into multiple subsequences. The number of similar students on the ASSIST10, ASSIST15, and EdNet datasets was set to 15, 15, and 10, respectively. The attention head for the common knowledge state and the expected answering response for the target exercise was 6.

CoKT: Improving model performance by measuring the similarity of students’ learning behaviors and utilizing inter-student information.

GKT: Learning relationships between knowledge concepts and formulating knowledge tracing as a time-series node-level classification problem for GCN.

AKT: Modeling students’ learning sequences using Transformer and proposing a monotonic attentional mechanism that optimizes the attentional weights for context.

Also for fairness, this paper uses equal amount of information input for all models. The performance comparison of different methods on three datasets is shown in Table 1. We can observe that the performance of DKT-FC outperforms the recently proposed deep learning based methods in terms of AUC and ACC metrics. More precisely, the AUC on ASSIST10, ASSIST15, and EdNet increases by 17.55%, 18.51%, and 14.21% over the three compared models, and the ACC of DKT-FC increases by 12.15%, 20.19%, and 19.5%, respectively. This implies that the introduction of collaborative information and self-supervised learning into DKT-FC favors its performance.

Table 1.

Different methods are compared in three data sets

Method	ASSIST10		ASSIST15		EdNet
Method	ACC	AUC	ACC	AUC	ACC	AUC
GKT	0.8535	0.8505	0.7842	0.7665	0.7714	0.7968
AKT	0.8001	0.7941	0.7906	0.8032	0.7804	0.8279
CoKT	0.8475	0.8647	0.7905	0.8103	0.7944	0.7994
DKT-FC	0.9216	0.9696	0.9861	0.9516	0.9664	0.9389

2.2.2

Ablation experiments

To further investigate and probe the roles and contributions of the various components of the DKT-FC model, we conduct relevant ablation experiments.

w/o collaborative information: Remove the collaborative information used in the DKT-FC model and learn the knowledge state only from the student’s personal interaction history information.

w/o self-supervised learning: Remove the self-supervised learning module from the DKT-FC model. Integrate only personal knowledge state and ensemble information and use the integrated information for prediction.

w/o ensemble information: Remove the module of integrating personal and ensemble information from the DKT-FC model. Only collaborative information is utilized as well as learning a good knowledge state representation through the use of self-supervised learning, and does not transform collaborative information into prediction probabilities.

The performance of the key components of the DKT-FC model is shown in Table 2.

Table 2.

Performance of key components of DKT-FC model

Serial number	Method	ASSIST10	ASSIST15	EdNet
0	DKT-FC(full model)	0.9225	0.9103	0.9281
1	w/o collaborative information	0.8368	0.8833	0.8245
2	w/o self-supervised learning	0.8572	0.8998	0.8495
3	w/o ensemble information	0.8858	0.8952	0.8584

DKT-FC integrates information from the common knowledge states of similar students and expected responses to the target exercise as collaborative information to make collaborative predictions. Methods 2 and 3 introduce collaborative information using both explicit and implicit strategies, respectively, and both show significant performance improvements over Method 1, which uses only personal information. These results suggest that collaborative information is indeed beneficial for improving knowledge tracking. In addition, our two ways of using collaborative information can complement each other, and DKT-FC performs significantly better than methods that use collaborative information in a single way.

We conducted experiments on three public datasets to assess the impact of influencing the number of similar students K on the performance of the AUC metric. We change the number of similar students from 1 to 20 because too much information about similar students may introduce noisy information in the collaboration. The effect of the number of similar students, K, on the model performance is shown in Figure 1, where (a)-(c) represent the ASSIST10, ASSIST15, and EdNet datasets, respectively. As can be seen from the figure, the performance of CoSKT improves with increasing K until it reaches its peak, and CoSKT reaches its best performance at 20, 19 and 12.5 on the ASSIST10, ASSIST15 and EdNet datasets, respectively.

2.2.3

Case studies

To further investigate the effect of synergistic information on the correct prediction of the next exercise, we visualized the prediction of Student u on Exercise t + 1. A visualization of the prediction of Student u’s responses on Exercise e_t+1 is shown in Figure 2. Each square in the figure represents one interaction. The first row represents the target student u’s history of answering exercise $I_{t + 1}^{u}$ $$\mathcal{I}_{t + 1}^u$$ and the upcoming answer to exercise e_t+1. The remaining 10 rows are the answer sequences of his first 10 similar students retrieved through the methods in this paper. Each similar student’s block consists of a question ID (number), a concept annotation (color), and the status of the square (with or without an outer box).

From the individual information shown in the figure, it can be seen that target student u has answered four exercises on the same concepts as e_t+1, but he was only 55% accurate in answering these exercises. Therefore, with his individual information it is difficult to predict whether he will be able to answer e_t+1 correctly or not, whereas target student u gave the correct answer. On the other hand, we can observe that 75% similar students answered correctly. This indicates that the target student has a high probability of answering correctly e_t+1. By combining synergistic information with personal information, our DKT-FC predicts that the student has 88.63% probability of answering the exercise correctly. It can be seen that synergistic information is favorable for prediction when there is less individual information.

2.3

Design and Development of Intelligent Teaching and Learning System

2.3.1

Overall system design

The functional components of a complete intelligent tutoring platform should mainly contain: collecting process data from the learner during the teaching process, analyzing and calculating the learner’s keyword mastery and generating student portraits and course portraits, recommending appropriate learning resources for the student to fit his/her learning path, and generating evaluative reports. The student portrait refers to the application of the results of each test, questionnaire results and the process data generated during self-study, including the collection of the learner’s test duration and number of times, mastery assessment and process evaluation. The system is constructed from an engineering perspective, and its purpose is to present the students’ learning process concretely through engineering ideas, so as to achieve the analysis and control of the learning process and personalized education for the learners, with the aim of cultivating the comprehensive quality of the learners, so that they can continuously improve their comprehensive quality on the basis of completing the contents required by the curriculum, and lay a good foundation for entering the society or receiving higher education.

2.3.2

Performance analysis and architectural design of the system

1)

Demand analysis (1)

Registration and Login:

By using the registration function learners will get an exclusive personal account, which is used to log in to the smart tutorial system and manage, maintain and operate personal information. This button has the purpose of registering and logging in the account, changing the password, and filling in and modifying the personal data.

(2)

Modeling the curriculum

Match the knowledge keywords with KWA and also be able to set up, add as well as manage classes and learners in this course.

(3)

Managing the question bank

The main operation performed by the platform administrator on the question bank is to manage the questions contained in the question bank accordingly. The system administrator can manually or automatically add, delete, and modify topics, match and label topics with knowledge keywords, KWAs, and other attributes, and filter topics based on a variety of conditions.

(4)

Releasing Test Papers

Question paper publishing is one of the uses of providing exams to the learners by the teachers. Teachers can form exams in three ways: (a) sampling - by teachers who filter the questions according to the conditional attributes in the question bank to form a fixed exam; (b) selecting - by teachers who manually add questions and then the system automatically generates the exam paper; (c) Random - set the rules in advance to form a test paper with non-fixed test questions, which are randomly formed each time a test paper is released, so that each learner has a unique and exclusive test paper.

(5)

Selection of Courses

Selection of courses is a learner’s choice of a particular course offered by a particular teacher when entering the student-side platform after registering and logging into the system. Selecting a course is an important prerequisite for evaluating the overall quality of students and making personalized recommendations on topics. Course selection includes two functions: selecting a course and switching courses.

(6)

Examination

Candidates can complete the questions according to their own choices within the stipulated time. If the candidates have not completed the paper by the end time, the system will automatically submit the paper. After completing the submission, the system will automatically adjudicate and grade the set questions, and finally the results of this test will be displayed in the system at a specific time.

(7)

Evaluation Report

The overall evaluation contains learners’ basic personal data, such as learners’ names, student numbers, classes, majors, colleges, etc.; topic information data, and data such as student portraits. Learners can view the evaluative report given by the system after each quiz at any time and from anywhere, or by pooling multiple quiz selections together in order to access the overall evaluation.

(8)

Viewing Results

Teachers can use the Results Lookup interface to see the data related to the quiz, specifically, the test scores, passes and failures, and the answer history for each candidate based on the candidate list.

(9)

Course Portrait

Learners can view their profile data through the Course Profile. By accessing and understanding the profile, learners can easily and intuitively see the level of their knowledge keywords and the results of mastery assessment.

2)

Architecture Design

After analyzing the requirements of the intelligent teaching and learning system, this paper adopts the Browser/Server architecture. The teacher management interface and student operation interface are independent of each other, and the application data are provided through the API application programming interface.

2.4

Innovative STSE-based teaching models

According to the development of inquiry-based learning theory, this study argues that the “5E” teaching model includes the main links of inquiry-based learning theory and embodies the main ideas of inquiry-based learning, and the “5E” teaching model includes five links: attraction, inquiry, explanation, transfer and evaluation. Based on the 5E teaching model and combined with STSE education theory, this study proposes an inquiry-based teaching model based on STSE education. The inquiry-based teaching model of higher education based on STSE education is shown in Figure 3.

1)

Linking elements of STSE education and creating authentic contexts

In the inquiry-based teaching model based on STSE education, students’ interest and motivation are stimulated through the creation of specific contexts linking science, technology, society and the environment, so that students can apply the knowledge and skills they have learned to solve practical problems in a real environment. The specific contents are as follows: (1)

Activate students’ old knowledge implementation strategy:

Teachers can present an engaging problem or a real-life situation in class to arouse students’ thinking and discussion. Teachers can take into account students’ daily life and social realities and guide them to recall relevant experiences and knowledge. Teachers can use multimedia resources or real-life case presentations to arouse students’ interest and awareness.

(2)

Teachers clarify theme implementation strategies

Teachers can design specific learning tasks or challenges to highlight the focus and objectives of learning. Teachers should provide clear learning instructions and objectives so that students understand the meaning and purpose of learning. Teachers can introduce real cases or situations so that students can understand the importance and application scenarios of the learning theme in practice.

2)

Leading to Information Knowledge and Exploring Technical Essentials

In the inquiry-based teaching model based on STSE education, independent inquiry is an important part of cultivating students’ active learning and problem-solving skills. Through independent inquiry, students will be able to explore topics of interest in depth, think independently and solve problems. (1)

Implementation Strategies for Students’ Independent Learning

Teachers design open-ended questions or tasks that require students to conduct independent inquiry and research. Teachers provide students with the required learning resources and instructional materials, such as books, online resources, and lab tools.

(2)

Student cooperative inquiry implementation strategy

Students form groups and work together under the guidance of the teacher to formulate research objectives, divide the work, collect and organize information, conduct experiments and observations, analyze data, and discuss and propose solutions together. Teachers encourage students to communicate and share ideas with each other in the process of cooperative inquiry.

3)

Core Knowledge Essentials, Key Techniques Points

This step is an important part to help students understand and master knowledge. Through the teacher’s explanation, students can understand the knowledge structure more deeply and systematically, and promote their application and extension of what they have learned. It is based on two activity links: teacher’s fine-tuning knowledge and teacher’s key pointing. (1)

Teacher’s fine-tuning knowledge implementation strategy

Teachers explain relevant concepts, principles and techniques through clear and concise language and examples. In the process of explanation, teachers should guide students to think and ask questions in order to activate their thinking and stimulate their curiosity.

(2)

Teachers’ Key Points Implementation Strategies

Teachers should focus on students’ confusions and problems and make key point analysis in time. Teachers can help students connect what they have learned with real problems by introducing real cases and application scenarios.

4)

Real Situation Application, Digital Competency Migration

Migration and application is a key link in the inquiry-based teaching model based on STSE education, which aims to cultivate students to apply what they have learned to real situations and other subject areas, and to promote the development of students’ comprehensive ability and innovative thinking. (1)

Real Situation Application Implementation Strategies

Teachers can design authentic cases or simulated scenarios for students to apply their acquired knowledge and skills in simulated real-life situations. Students can engage in authentic contextual application by participating in project-based learning. Teachers can organize role-playing and practical exercises for students to practically apply knowledge and skills in simulated situations.

(2)

Extended Transfer Application Implementation Strategies

Students can participate in interdisciplinary integration projects to combine knowledge with other subject areas to solve interdisciplinary problems or carry out interdisciplinary projects. Students can expand the application scenarios of digital competency migration through inquiry-based learning and independent projects. Teachers can organize students to participate in industry practice or social application projects, so that they can apply the knowledge and skills they have learned in real work or social practice.

5)

Introspection of Knowledge and Skills, Summarization and Evaluation

At this stage, teachers should guide students to review the knowledge and skills they have acquired and motivate them to organize what they have learned into a logical and clear framework for better understanding and mastery of what they have learned. The process of organizing knowledge and skills helps students to gain a deeper understanding of the concepts and principles of what they have learned, and to connect fragmented knowledge points to form a more complete body of knowledge. (1)

Reflective Self-Knowledge Implementation Strategy

Students should be encouraged to conduct in-depth self-evaluation and reflection on their learning process. Students can record their learning process and insights by writing study notes and journals. Peer communication and sharing among students is also an effective way of reflection.

(2)

Summarizing Assessment Implementation Strategies

Teachers can design some comprehensive assessment tasks that require students to apply the knowledge and skills they have learned in real-life situations so as to test their comprehensive application skills. When conducting summary assessment, teachers should clarify the assessment criteria so that students clearly understand the standards and requirements they are being assessed. Teachers should give students timely feedback and guidance to help them identify their shortcomings and provide suggestions and methods for improvement.

3

Analysis of the results of students’ learning effectiveness under the innovative teaching model

This paper takes 352 students majoring in Chinese language from freshmen to seniors in a university in City B as the research subjects, which has sufficient and comprehensive faculty and teaching equipment, and the Chinese language course is one of the basic courses in this school, so it has sufficient experimental conditions. In order to exclude the influence of the lecturer on the experimental results, the content of the two tests is the same. In this paper, 352 students were selected from the class taught by the same teacher. It is divided into experimental group (166) and control group (186). The experimental group adopts the innovative teaching mode of intelligent teaching aids based on DKT-FC, and the control group adopts the traditional teaching mode.

3.1

Descriptive statistical results

The means and standard deviations of the first reading tally scores of the experimental and control groups are shown in Table 3. The experimental group’s first reading tally scores were lower than the control group’s overall and in all dimensions.

Table 3.

The mean and standard deviation of the first reading

Dimension	Group	N	Mean value	Standard deviation	Standard error mean
Integral summary 1	Experimental group	166	2.7908	0.6989	0.0807
Integral summary 1	Control group	186	3.2801	0.7012	0.0829
Continuum elaboration 1	Experimental group	166	3.0685	0.7001	0.0806
Continuum elaboration 1	Control group	186	3.386	0.7027	0.0785
Integral connection 1	Experimental group	166	2.633	0.7006	0.081
Integral connection 1	Control group	186	3.1312	0.7012	0.0793
General question 1	Experimental group	166	1.6656	0.7	0.0808
General question 1	Control group	186	2.2775	0.6993	0.0786
Total score 1	Experimental group	166	10.1579	2.7996	0.3231
Total score 1	Control group	186	12.0748	2.8044	0.3193

The means and standard deviations of the second reading tally scores of the experimental and control groups are shown in Table 4. The experimental group’s second reading tally scores were lower than the control group’s overall and in all dimensions.

Table 4.

The mean and standard deviation of the second reading

Dimension	Group	N	Mean value	Standard deviation	Standard error mean
Integral summary 2	Experimental group	166	3.193	0.5986	0.0706
Integral summary 2	Control group	186	3.3279	0.5967	0.0701
Continuum elaboration 2	Experimental group	166	3.1252	0.5988	0.0708
Continuum elaboration 2	Control group	186	3.3597	0.5992	0.0697
Integral connection 2	Experimental group	166	2.9476	0.6004	0.0722
Integral connection 2	Control group	186	3.2041	0.6006	0.0695
General question 2	Experimental group	166	2.1356	0.6004	0.07
General question 2	Control group	186	2.3767	0.6004	0.0712
Total score 2	Experimental group	166	11.4014	2.3982	0.2836
Total score 2	Control group	186	12.1955	2.4975	0.2903

3.2

Analysis of covariance

The results of the analysis of the first reading integration show that there is a significant difference between the reading integration ability of the experimental group and the control group, which is likely to be related to the influence of the students’ own level of integration ability. Therefore, this paper takes the achievement of the first reading integration as a covariate and uses the analysis of covariance to compare and analyze the students’ achievement of the second reading integration. The second reading tally-summary main effect test is shown in Table 5. The DKT-FC-based innovative teaching model of smart tutoring has a factor of influence on the summary dimension of second reading integration (P=0). And the summary dimension of the second reading tally is significantly insignificant (P=0.568 > 0.05). Whereas the first experiment’s unified summary significantly influences the results of the second experiment (P < 0.05).

Table 5.

The second reading is the main effect test

Source	Class sum of Ⅲ	Freedom	Mean square	F	Sig.	Eta squared	Non-central parameter
Modified model	25.0309a	2	11.974	29.4656	0	0.3049	59.1283
Intercept	31.9934	1	31.9934	77.0167	0	0.3006	77.0167
Group	0.2047	1	0.2047	0.3831	0.568	0	0.3831
Integral summary 1	23.9134	1	23.9134	57.0643	0	0.2445	57.0643
Error	70.9653	350	0.4105
Total	1929.6482	352
Revised total	95.8728	351

The estimated values of the second reading tally-summary parameters are shown in Table 6. As can be seen from the table, after excluding the effect of the first reading integration summary data, the data of the experimental group increased by 0.0383 points compared to the control group, but the difference was not significant (p=0.568), which says that there is no significant difference in the ability to summarize in the second reading integration.

Table 6.

The second reading system is the estimation value of the parameter

Parameter	B	Standard error	t	Sig.	95% confidence interval		Eta squared
Parameter	B	Standard error	t	Sig.	Lower limit	Upper limit	Eta squared
Intercept	1.7614	0.2007	8.0018	0	1.2895	2.119	0.2565
Group=1	0.0383	0.0977	0.3981	0.568	-0.1466	0.2236	0.001
Group=2	0a	—	—	—	—	—	—
Integral summary 1	0.4815	0.0608	7.5012	0	0.2886	0.6174	0.1765

The second reading integration-elaboration main effect test is shown in Table 7. The innovative teaching model of smart tutorials based on DKT-FC has influencing factors on the second reading integration-elaboration dimension but the difference is not significant (P=0.4878), while the data of the first experiment significantly influences the data of the second experiment (P<0.05).

Table 7.

The second reading is explained by the main effect test

Source	Class sum of Ⅲ	Freedom	Mean square	F	Sig.	Eta squared	Non-central parameter
Modified model	23.9912a	2	11.9989	24.0012	0	0.2205	47.9943
Intercept	27.9087	1	27.9087	55.9991	0	0.2576	55.9991
Group	0.1752	1	0.1752	0.3017	0.4878	0.0012	0.3017
Continuum elaboration 1	23.0243	1	23.0243	45.001	0	0.2046	45.001
Error	85.9998	350	0.5013
Total	2050.0154	352
Revised total	111.048	351

The estimated values of the second reading integration-elaboration parameters are suffered by Table 8. After excluding the effect of the first reading integration elaboration data, the adjusted data of the experimental group was 0.0504 points lower than the adjusted data of the control group, but the difference was not significant (0.4878), which shows that excluding the effect of the first reading integration elaboration, there was no significant difference between the elaboration ability of the experimental and control groups in the second reading integration.

Table 8.

The second reading is complete - the parameter estimation is described

Parameter	B	Standard error	t	Sig.	95% confidence interval		Eta squared
Parameter	B	Standard error	t	Sig.	Lower limit	Upper limit	Eta squared
Intercept	1.7972	0.2504	7.2006	0	1.3022	2.3082	0.1865
Group=1	-0.0504	0.0995	-0.5711	0.4878	-0.2477	0.1343	0.002
Group=2	0a	—	—	—	—	—	—
Continuum elaboration 1	0.4788	0.0703	6.7012	0	0.3054	0.6083	0.1955

The second reading integration-linkage main effect test is shown in Table 9. The DKT-FC-based innovative teaching model of smart tutorials had an influential factor on the second reading integration-connection dimension (P=0<0.001). There was no significant effect of group on the linking dimension of second reading integration (P=0.3601). The data from the first experiment significantly influenced the data from the second experiment (P=0<0.05).

Table 9.

The second reading is tested by the main effect

Source	Class sum of Ⅲ	Freedom	Mean square	F	Sig.	Eta squared	Non-central parameter
Modified model	25.7961a	2	12.8998	26.0003	0	0.2368	53.0088
Intercept	46.9965	1	46.9965	96.9982	0	0.3681	96.9982
Group	0.3922	1	0.3922	0.8011	0.3601	0.0136	0.8011
Integral connection 1	21.9977	1	21.9977	45.997	0	0.187	45.997
Error	80.9904	350	0.4811
Total	1764.9947	352
Revised total	106.9802	351

The estimated values of the second reading integration-linkage parameters are shown in Table 10. As can be seen from the table, the experimental group’s adjusted data was 0.087 points lower than the control group’s adjusted data (P=0.3601), which indicates that there was no significant difference between the experimental and control groups’ ability to link in the second reading integration after excluding the effect of the first reading integration elaboration.

Table 10.

The second reading integer - the link parameter estimation

Parameter	B	Standard error	t	Sig.	95% confidence interval		Eta squared
Parameter	B	Standard error	t	Sig.	Lower limit	Upper limit	Eta squared
Intercept	1.8961	0.2001	9.1978	0	1.4861	2.3187	0.3557
Group=1	-0.087	0.1076	-0.8984	0.3601	-0.2827	0.1083	0.005
Group=2	0a	—	—	—	—	—	—
Integral connection 1	0.3702	0.0596	6.8024	0	0.2972	0.5249	0.1967

The second reading integration-questioning main effect test is shown in Table 11. There are factors in the DKT-FC-based innovative teaching model of smart tutoring that have an effect on the second reading integration-questioning dimension (P < 0.05), but there is no significant effect on the second reading integration-questioning dimension (P = 0.403 > 0.05). The data from the first experiment significantly influenced the data from the second experiment (P = 0 < 0.05).

Table 11.

The second reading is the main effect test

Source	Class sum of Ⅲ	Freedom	Mean square	F	Sig.	Eta squared	Non-central parameter
Modified model	15.0101a	2	7.0018	12.9978	0	0.1167	26.9912
Intercept	84.0071	1	84.0071	149.0034	0	0.4695	149.0034
Group	0.1986	1	0.1986	0.4705	0.403	0.0098	0.4705
General question 1	11.9914	1	11.9914	21.9993	0	0.1185	21.9993
Error	94.0148	350	0.5629
Total	1069.0153	352
Revised total	109.0058	351

The estimated values of the second reading-unification-questioning parameters are shown in Table 12. The adjusted data of the experimental group was 0.0631 points lower than the adjusted data of the control group (P > 0.05), which indicates that excluding the effect of the first reading integration questioning, there was no significant difference between the experimental and control groups in their ability to ask questions in the second reading integration.

Table 12.

The second reading is an integer - question parameter estimation

Parameter	B	Standard error	t	Sig.	95% confidence interval		Eta squared
Parameter	B	Standard error	t	Sig.	Lower limit	Upper limit	Eta squared
Intercept	1.7023	0.1602	9.9976	0	1.4721	1.9925	0.424
Group=1	-0.0631	0.1187	-0.6786	0.403	-0.2727	0.1562	0.003
Group=2	0a	—	—	—	—	—	—
Integral summary 1	0.3939	0.0599	4.6951	0	0.1941	0.3753	0.1272

The main effect test for the second reading tally-total score is shown in Table 13. There were factors that had an effect on the second reading unity-total score based on the DKT-FC’s innovative teaching model of smart tutoring (P=0). There is no significant effect of the unified-total score on the second reading unified total score (P=0.6987). The first experimental data significantly affects the second experimental data (P < 0.05).

Table 13.

The second reading is tested for the main effect

Source	Class sum of Ⅲ	Freedom	Mean square	F	Sig.	Eta squared	Non-central parameter
Modified model	36.9919a	2	162.9993	64.0007	0	0.4052	129.0152
Intercept	186.0073	1	186.0073	72.9993	0	0.2941	72.9993
Group	0.2671	1	0.2671	0.1023	0.6987	0.0193	0.1023
Total score 1	297.9945	1	297.9945	117.0002	0	0.1556	117.0002
Error	424.9919	350	2.5003
Total	25899.993	352
Revised total	753.0152	351

The estimated values of the second reading tally-total score parameters are shown in Table 14. After excluding the effect of the first reading integration-total score, the adjusted data of the experimental group is 0.0913 points higher than the adjusted data of the control group, but there is no significance (P=0.6987>0.05), which can be seen that after excluding the effect of the first reading integration-total score, there is no significant difference between the experimental group and the control group’s total score of the second reading integration.

Table 14.

The second reading integer - the total score parameter estimation

Parameter	B	Standard error	t	Sig.	95% confidence interval		Eta squared
Parameter	B	Standard error	t	Sig.	Lower limit	Upper limit	Eta squared
Intercept	5.3053	0.5981	7.7988	0	3.9853	6.7041	0.2578
Group=1	0.0913	0.501	0.2981	0.6987	-0.3851	0.6004	0.0282
Group=2	0a	—	—	—	—	—	—
Total score 1	0.5916	0.0492	10.7971	0	0.3863	0.7096	0.3854

From the above conclusion, it can be seen that teaching text reading under the innovative teaching mode can make the experimental group’s reading integration ability higher than that of the control group, but the difference between the two groups is not significant.

3.3

Correlation analysis

The correlations of the dimensions of reading integration in the experimental group are shown in Table 15. As can be seen from the table, all the correlations between all the indicators of the students in the experimental group showed significant positive correlations (P < 0.05). Among them, the correlation between the integrated total score and the integrated elaboration is the largest, while the correlation between the integrated link and the integrated summary is the smallest, with correlation coefficients of 0.8579 and 0.2375, respectively. It can be seen that all dimensions within the experimental group’s reading integration are positively correlated with each other.

Table 15.

The experimental group is concerned with the dimensional correlation

Dimension	Integral summary	Continuum elaboration	Integral connection	General question	Total score
Integral summary	1
Continuum elaboration	0.5526**	1
Integral connection	0.2375**	0.6192**	1
General question	0.3243**	0.368**	0.3103**	1
Total score	0.6863**	0.8579**	0.7505**	0.6709**	1

The correlations between the dimensions of reading integration in the control group are shown in Table 16. In the control group, there was a positive correlation between the dimensions of students’ reading integration, but the correlations varied widely. Specifically, the difference between the co-ordinated questioning and the co-ordinated summary and co-ordinated elaboration was not significant, with correlation coefficients of only 0.0614 and 0.14 (p > 0.05). In addition, the correlation between the co-ordination total score and co-ordination elaboration was still the largest in the control group (0.703), but the correlation between the indicators was generally lower in the control group compared to the experimental group. This shows that students’ learning is more effective under the innovative teaching model using the DKT-FC-based intelligent teaching aid.

Table 16.

Control group read the whole dimension correlation

Dimension	Integral summary	Continuum elaboration	Integral connection	General question	Total score
Integral summary	1
Continuum elaboration	0.3113**	1
Integral connection	0.3041**	0.4003**	1
General question	0.0614	0.14	0.2494*	1
Total score	0.6261**	0.698**	0.703**	0.4785**	1

4

Conclusion

In this paper, a deep knowledge tracking model (DKT-FC) is constructed to monitor students’ learning by combining the Ebbinghaus forgetting curve, on the basis of which an intelligent tutorial system is constructed to master the students’ learning dynamics, and finally, the effectiveness of the application of the innovative teaching model of STSE is verified by the covariance and correlation analyses. The AUC and ACC of this paper’s model on the three benchmark datasets of ASSIST10, ASSIST15 and EdNet increased by 14.21%-18.51% and 12.15%-20.19%, respectively, compared with the three comparative models, and it is obvious that the introduction of collaborative information and self-supervised learning into DKT-FC favors its performance. It can be seen that the overall performance of the DKT-FC model proposed in this paper, as well as the performance of each module, is outstanding, and has a certain degree of effectiveness and wide applicability. The use of the DKT-FC-based innovative teaching model of intelligent tutoring can make the correlation between the total score of co-ordination and co-ordination elaboration of the experimental group of students generally higher than that of the control group, which in turn effectively improves the learning effectiveness of students and promotes the overall development of students.

Language:: English

Publication timeframe:: 1 times per year
Journal Subjects:: Life Sciences, Life Sciences, other, Mathematics, Applied Mathematics, General Mathematics, Physics, Physics, other

Journal RSS Feed

Analysis of Teaching Mode Innovation and Learning Effectiveness Assisted by Artificial Intelligence

Haiying Luo

Yanhong Wu

Published Online: Sep 26, 2025

Received: Jan 18, 2025

Accepted: May 06, 2025

DOI: https://doi.org/10.2478/amns-2025-1071

KeywordsDeep Knowledge Tracking Model (DKT-FC), STSE, Analysis of Covariance, Artificial Intelligence

© 2025 Haiying Luo et al., published by Sciendo

This work is licensed under the Creative Commons Attribution 4.0 International License.

Keywords
Deep Knowledge Tracking Model (DKT-FC), STSE, Analysis of Covariance, Artificial Intelligence